Bearing fault recognition method based on lifting one-dimensional local ternary pattern

Abstract

The application belongs to the technical field of fault diagnosis, and specifically discloses a bearing fault recognition method based on a lifting one-dimensional local ternary pattern, which comprises the following steps: firstly, obtaining the vibration signal of the equipment where the bearing is located, and calculating the first-order differential signal and the second-order differential signal thereof; secondly, constructing a multiple cyclic period matrix containing the vibration signal and the two types of differential signals according to the period point number of the signal; thirdly, quantizing the matrix by using the lifting one-dimensional local ternary pattern with the local root mean square of the signal in the window as the quantization criterion and reconstructing the matrix into a decimal matrix; fourthly, calculating the evaluation index value of each row signal of the decimal matrix and constructing a high-dimensional feature set; and finally, inputting the high-dimensional feature set into a model that is pre-trained by using a pattern recognition algorithm to realize bearing fault classification and recognition. The application solves the problems of inaccurate quantization and incomplete feature extraction of the traditional one-dimensional local ternary pattern method, can accurately mine the bearing fault features, and improves the accuracy and reliability of fault recognition.

Description

Technical Field

[0001] This invention belongs to the field of fault identification technology, and specifically relates to a bearing fault identification method based on an enhanced one-dimensional local ternary mode. Background Technology

[0002] Bearings, often referred to as "the joints of industry," are widely used in rotating machinery. Due to the harsh working environment of such equipment, bearings are prone to damage, and their health directly affects the stable operation of the equipment. Therefore, condition monitoring and fault diagnosis of bearings are crucial.

[0003] In existing bearing fault identification technologies, time-frequency domain feature extraction methods are crucial, including signal decomposition algorithms such as wavelet transform, empirical mode decomposition, and variational mode decomposition, as well as methods based on the cyclic periodicity of vibration signals, such as cyclic autocorrelation and cyclic stationarity analysis. In recent years, scholars have attempted to identify bearing conditions from the perspective of local feature extraction using local binary pattern (LBP) and one-dimensional local binary pattern (1D-LBP). However, 1D-LBP only has two modes, 0 and 1, which does not comprehensively characterize the internal features of the vibration signal.

[0004] To address this issue, some scholars have proposed a one-dimensional localternary pattern (1D-LTP), which uses additional operators and upper and lower thresholds to achieve a more accurate description of the internal features of one-dimensional vibration signals. However, the classic 1D-LTP quantizes signals based on local intermediate values, failing to consider the characteristics of the rotating machinery itself and the vibration signals during bearing failures. This results in limitations such as inaccurate quantization and an inability to fully extract local signal features, making it unsuitable for bearing fault identification under multi-load conditions. Summary of the Invention

[0005] In view of this, the present invention proposes a bearing fault identification method based on a one-dimensional local ternary mode to solve the problems existing in the prior art.

[0006] This invention provides a bearing fault identification method based on a one-dimensional local ternary mode, comprising:

[0007] S1: Collect vibration signals from the equipment where the bearing is located. , , For data length;

[0008] S2: Calculate the vibration signal First-order difference signal and second-order differential signals ;

[0009] S3: Based on the vibration signal First-order differential signal and second-order differential signals Constructing a multi-cycle periodic matrix ,in, , , The number of columns is the vibration signal Points in one cycle or Integer multiples of;

[0010] S4: Based on the enhanced one-dimensional local ternary mode, the multi-cyclic periodic matrix is... Perform ternary quantization and reconstruct the quantization result into a decimal matrix. The one-dimensional local ternary mode is improved by using the local root mean square of the signal within the window as the quantization criterion.

[0011] S5: Calculate the decimal matrix Evaluation index value of each line of signal Construct a high-dimensional feature set;

[0012] S6: Reduce the dimensionality of the high-dimensional feature set and construct a feature vector based on the reduced feature set;

[0013] S7: Input the feature vector into a model pre-trained using a pattern recognition algorithm to achieve classification and recognition of bearing status.

[0014] Preferably, in S2, the vibration signal First-order difference signal and second-order differential signals The calculation formula is as follows:

[0015] ;

[0016] or ;

[0017] ;

[0018] or: .

[0019] Further preferably, in S3, the vibration signal Periodic points , To round up, For rotational speed, The sampling frequency.

[0020] Further optimization, in S3, ;

[0021] ;

[0022] ;

[0023] In the formula, The number of cycles, , This indicates rounding up to the nearest integer.

[0024] Further optimization, in S4, in the enhancement of the one-dimensional local ternary mode, the rectangular window is set to... With a sliding step size of 1, starting from the initial position, the eight signals within the window, excluding the center point, are ternarily quantized.

[0025] Further optimization, in S4, based on the enhanced one-dimensional local ternary mode, is applied to the eight signals within the window excluding the center point. Perform ternary quantization to obtain the quantized signal. The formula is as follows, where, :

[0026] ;

[0027] in, for The root mean square of; for The standard deviation.

[0028] Further optimization, in S4, the quantized signal... Convert to decimal sequence The formula is as follows:

[0029] .

[0030] Further optimization, in S5, the evaluation index value is the fuzzy entropy value.

[0031] In a further preferred embodiment, in S6, the dimensionality reduction method is a joint approximate diagonalization matrix method.

[0032] In a further preferred embodiment, in S7, the pattern recognition algorithm is a support vector machine or a deep learning algorithm.

[0033] This invention proposes a bearing fault identification method based on enhanced one-dimensional local ternary mode. It constructs a multi-cycle periodic matrix using vibration signals and their first-order and second-order difference signals to comprehensively capture the characteristic information (impact and periodic features) of bearing faults, solving the problem of incomplete feature extraction in traditional methods. It improves the classic 1D-LTP quantization criterion by replacing local intermediate values ​​with root mean square (RMS), which is more sensitive to the impact features of bearing faults. This fully considers the inherent characteristics of the vibration signals during bearing faults, improving quantization accuracy and adapting to various load conditions. Combining fuzzy entropy features and a joint approximate diagonalization dimensionality reduction method, it achieves effective extraction and compression of high-dimensional features. Finally, it uses a pattern recognition algorithm to achieve accurate classification of bearing faults. Experimental results show that this invention significantly improves fault identification accuracy compared to traditional methods, demonstrating good practicality and reliability. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart of the bearing fault identification method based on enhanced one-dimensional local ternary mode provided by the present invention;

[0036] Figure 2 This is a schematic diagram of signal quantization based on improving a one-dimensional local ternary mode;

[0037] Figure 3 The images show the recognition results obtained by different methods, where (a) is the recognition result image of comparison method 1, (b) is the recognition result image of comparison method 2, and (c) is the recognition result image of the method proposed in this invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, to avoid obscuring the invention with unnecessary details, only processing steps closely related to the solution of this invention are shown in the drawings, while other details not closely related to this invention are omitted.

[0039] like Figure 1 As shown, this invention provides a bearing fault identification method based on a one-dimensional local ternary mode, comprising:

[0040] S1: Collect vibration signals from the equipment where the bearing is located. , , For data length;

[0041] S2: Calculate the vibration signal First-order difference signal and second-order differential signals ;

[0042] Among them, vibration signal First-order difference signal and second-order differential signals The calculation formula is as follows:

[0043] ;

[0044] or ;

[0045] ;

[0046] or: .

[0047] S3: Based on the vibration signal First-order differential signal and second-order differential signals Constructing a multi-cycle periodic matrix ,in, , , The number of columns is the vibration signal Points in one cycle or Integer multiples of;

[0048] Among them, the vibration signal Periodic points , To round up, For rotational speed, The sampling frequency;

[0049] ;

[0050] ;

[0051] ;

[0052] In the formula, The number of cycles, , Indicates rounding down;

[0053] S4: Apply the Improved One-Dimensional Local Ternary Mode (I1D-LTP) method to the multi-cyclic periodic matrix. Perform ternary quantization and reconstruct the quantization result into a decimal matrix. The one-dimensional local ternary mode is improved by using the local root mean square of the signal within the window as the quantization criterion.

[0054] Among them, such as Figure 2 As shown, in the enhanced one-dimensional local ternary mode, the rectangular window is set as... With a sliding step size of 1, starting from the initial position, the eight signals within the window, excluding the center point, are ternarily quantized.

[0055] Based on the enhanced one-dimensional local ternary mode (I1D-LTP), the eight signals within the window, excluding the center point, are analyzed. Perform ternary quantization to obtain the quantized signal. ( The formula for () is as follows:

[0056] ;

[0057] in, for The root mean square of; for Standard deviation;

[0058] quantized signal Convert to decimal sequence The formula is as follows:

[0059] ;

[0060] S5: Calculate the decimal matrix Evaluation index value of each line of signal Construct a high-dimensional feature set, and optimize it. The evaluation index value is the fuzzy entropy value, which can also be the average value, variance, information entropy, or other index values.

[0061] S6: Reduce the dimensionality of the high-dimensional feature set and construct a feature vector based on the reduced feature set. Preferably, the high-dimensional feature set is reduced using a joint approximate diagonalization feature matrix method.

[0062] S7: Input the feature vector into a model pre-trained using a pattern recognition algorithm to achieve classification and recognition of bearing status. The pattern recognition algorithm is preferably a support vector machine (SVM) or a deep learning algorithm.

[0063] Example 1

[0064] The effectiveness of the bearing fault identification method based on the improved one-dimensional local ternary mode proposed in this invention is verified below using the drive-end acceleration sensor data from the Case Western Reserve University bearing dataset as an example.

[0065] 1. Comparison method settings:

[0066] Comparison Method 1: Directly calculate the kurtosis value of the vibration signal and construct a feature set for classification and recognition;

[0067] Comparison Method 2: Using the classic 1D-LTP quantized signal (with local intermediate values ​​as the quantization criterion), the fuzzy entropy value of the decimal sequence is calculated and a feature set is constructed for classification and recognition.

[0068] Both comparison methods are based on the same data.

[0069] 2. Data preparation:

[0070] Vibration signals were acquired for bearings under normal conditions and with different damage levels (0.007mm, 0.014mm, 0.021mm) and loads (0hp, 1hp, 2hp, 3hp) for inner ring, outer ring, and rolling element faults, totaling 10 states. The vibration data were expanded using sliding window segmentation with a window length of 2048 and a step size of 1024. 100 sets of signal data were collected for each state (totaling 3*10*100=3000 sets of data). For each state, 50% of the data was randomly selected as the training set, and the remaining 50% as the test set. Support Vector Machine (SVM) was used as the classification algorithm. Detailed descriptions of the fault data are shown in Table 1.

[0071] Table 1. Detailed Description of Data

[0072]

[0073] 3. Experimental Results:

[0074] Comparison methods 1-2 with the method of the present invention are shown in the following results. Figure 3 As shown in Table 2. Figure 3 The horizontal axis represents samples of each state, and the vertical axis represents the ten operating states of the bearing, corresponding to Nor, IRF1, IRF2, IRF3, BF1, BF2, BF3, ORF1, ORF2, and ORF3 in Table 1. (The blue lines indicate these states.) For the actual category, red. The predicted category is shown in Table 2, which presents the average and standard deviation of the recognition rates for each method across ten consecutive random tests.

[0075] Table 2. Mean and standard deviation of recognition rate in ten consecutive random tests

[0076]

[0077] analyze Figure 3 As can be seen from Table 2:

[0078] Figure 3 (a) and in Table 2, compared to Method 1 (kurtosis in the classic method), the standard deviation value is the largest, the stability is poor, and the recognition rate is not high; Figure 3 (b) Compared with Method 2 in Table 2 (when the classic ternary quantization method is combined with fuzzy entropy), the standard deviation is smaller than that of the kurtosis comparison method; however, this method has more severe overlap between different states, which leads to a decrease in recognition accuracy. Figure 3 (c) The method proposed in Table 2 of this invention, because it considers the inherent characteristics of rotating machinery and the impact characteristics exhibited by the signal during bearing failure before quantization, improves the possibility of fault information extraction by quantizing the constructed "multiple cyclic periodic matrix" instead of the "original vibration signal". Simultaneously, when quantizing the signal according to the 1D-LTP method, it also considers the characteristic laws of the vibration signal during bearing failure, using the root mean square value that reflects the impact characteristics during bearing failure instead of local intermediate values ​​for quantization, thus achieving in-depth mining of fault information. Therefore, its state recognition effect is the most ideal.

[0079] Based on the above comparative analysis and classification results, it can be seen that the method proposed in this invention achieved a 100% recognition rate in ten random tests, and its recognition accuracy and stability are significantly better than other different comparative methods.

[0080] In summary, the method proposed in this invention solves the problems of inaccurate signal quantization and difficulty in accurately extracting bearing fault features in traditional feature extraction methods. Before quantizing the signal using the 1D-LTP method, the characteristics of the rotating machinery itself and the characteristic patterns of the signal during bearing faults are considered. A method that uses a constructed cyclic periodic matrix instead of the original signal for quantization is employed, thus highlighting the equipment's characteristic information. During signal quantization, a method that uses local root mean square values ​​that reflect the signal impact characteristics during bearing faults instead of local intermediate values ​​is used, enabling in-depth mining of local fault features. Therefore, the method proposed in this invention can accurately extract the features corresponding to bearing faults and achieve accurate classification of bearing fault types.

[0081] It should be noted that the purpose of disclosing the embodiments is to help further understand the present invention; however, those skilled in the art will understand that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments, and the scope of protection of the present invention is defined by the scope of the claims.

Claims

1. A bearing fault identification method based on a one-dimensional local ternary mode, characterized in that, include: S1: Collect vibration signals from the equipment where the bearing is located. , , For data length; S2: Calculate the vibration signal First-order difference signal and second-order differential signal ; S3: Based on the vibration signal First-order differential signal and second-order differential signal Constructing a multi-cycle periodic matrix ,in, , , The number of columns is the vibration signal Points in one cycle or Integer multiples of; S4: Based on the enhanced one-dimensional local ternary mode, the multiple cyclic periodic matrix is... Perform ternary quantization and reconstruct the quantization result into a decimal matrix. In this context, the one-dimensional local ternary mode uses the local root mean square of the signal within the window as the quantization criterion. In this one-dimensional local ternary mode, the rectangular window is set to... The sliding step size is 1. Starting from the initial position, it checks the 8 signals within the window, excluding the center point. Perform ternary quantization to obtain the quantized signal. The formula is as follows, where, : ； in, for The root mean square of; for Standard deviation; S5: Calculate the decimal matrix Evaluation index value of each line of signal Construct a high-dimensional feature set; S6: Reduce the dimensionality of the high-dimensional feature set and construct a feature vector based on the reduced feature set; S7: Input the feature vector into a model pre-trained using a pattern recognition algorithm to achieve classification and recognition of bearing status.

2. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S2, the vibration signal First-order difference signal and second-order differential signal The calculation formula is as follows: ； or ; ； or: .

3. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S3, the vibration signal Periodic points , To round up, For rotational speed, The sampling frequency.

4. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S3 ; ； ； In the formula, The number of cycles, , This indicates rounding up to the nearest integer.

5. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S4, the quantized signal Convert to decimal sequence The formula is as follows: 。 6. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S5, the evaluation index value is the fuzzy entropy value.

7. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S6, the dimensionality reduction method is the joint approximate diagonalization matrix method.

8. The bearing fault identification method based on a one-dimensional local ternary mode as described in claim 1, characterized in that, In S7, the pattern recognition algorithm is a support vector machine or a deep learning algorithm.

Images